In earlier slides, we saw that the +2 rows are necessary to ensure no dependency between chunks. The code here seems to suggest that Halide is smart enough to generate performant code with this kind of padding and the programmer doesn't need to worry about the exact implementation. Is this true in general, especially with increasing number of Halide functions in the pipeline (e.g. Google HDR+)?
In earlier slides, we saw that the +2 rows are necessary to ensure no dependency between chunks. The code here seems to suggest that Halide is smart enough to generate performant code with this kind of padding and the programmer doesn't need to worry about the exact implementation. Is this true in general, especially with increasing number of Halide functions in the pipeline (e.g. Google HDR+)?